03:36 imirkin: Lyude: there are, on g84+ gpus, subchan methods for waiting on fences
03:36 imirkin: (not sure if that answers your question)
07:18 orbea: in gentoo with mesa9999 I'm noticing wine-nine-standalone is broken and can't find any available backends. However this is only with the nouveau ddx + DRI2, it works with DRI3 + modesetting or nouveau and DRI2 + modesetting. Also xnine works everywhere...
07:20 orbea: it also works when starting the nouveau ddx with DRI3 and then using LIBGL_DRI3_DISABLE=true
13:48 imirkin: orbea: bisect?
13:48 imirkin: there were some timing changes
13:49 orbea: imirkin: occcurs with the stable package too, not sure how far back I would need to go
13:51 imirkin: oh, this is nine-only?
13:51 imirkin: nine kinda requires DRI3 iirc
13:52 imirkin: check with those guys
13:54 orbea: imirkin: xnine works as does modesetting + dri2
13:54 imirkin: ok
13:54 imirkin: well i have no clue.
13:54 imirkin: the nine people might.
13:54 orbea: i mentioned it to them, they thought it was wierd too
13:55 orbea: im also not sure how much longer I will have nouveau in that box, it was a temporary measure
13:55 imirkin: ok
21:03 orbea: the firefox right click menu has corrupted graphics with DRI3 + the nouveau ddx
21:03 orbea: modesetting + DRI3 works
21:03 orbea: I guess that is what I'll use
21:35 karolherbst: orbea: mhh, any idea what changed?
21:37 orbea: karolherbst: not sure, I just know I changed my xorg config from the nouveau ddx to modesetting to investigate a nine bug which it also avoided...
21:38 karolherbst: ahh
21:39 orbea: it just seems the nouveau ddx has various bugs...
21:40 orbea: i think i used modesetting when I was using this card last
21:52 ccr:rubs his head in wonder.
21:56 ccr: perhaps someone who understands nouveau kernel code could check this: in nouveau_gem.c : nouveau_gem_pushbuf_reloc_apply() u_free(reloc) is called, but then the same u_free(reloc) is done also in nouveau_gem_ioctl_pushbuf() which calls nouveau_gem_pushbuf_reloc_apply(). it would seem like a use-after-free/double free to me. kasan is not saying anything, but I'm not sure if I'm hitting the right code path.
21:59 ccr: previously the allocation and freeing of reloc was done entirely inside nouveau_gem_pushbuf_reloc_apply(), but 03e0d26fcf791e48164ff7c280c71225c361a89e changed it. to my casual observer eye it seems like the u_free() was left in by accident.